Featured Image

CTO Perspective: What to Know About AI Agents

Data Science and AI

By FactSet Insight  |  September 3, 2025

AI agents have emerged as a powerful force in reshaping financial workflows given they offer new possibilities for automation, efficiency, and scale. In this Q&A, FactSet Chief Technology Officer Kate Stepp shares a pragmatic view of where agents fit into the evolving AI landscape, and what business and technology leaders should be considering now.

Q: What are AI agents?

A: AI agents offer a leap forward in the evolution of artificial intelligence as systems. They work beyond the information retrieval and language-based tasks of GenAI to act on your behalf. Agents understand user goals, plan a course of action, interact with tools and data, iterate and reason, and complete tasks autonomously. They are aware of their environment and can learn and interact with other agents.

Q: What’s driving their prominence?

A: Agents are material because they are executing tasks at a speed, scale, and complexity level that was previously not feasible. In the past, automation was achieved through more rigid patterns and logic, replicating user behavior. This was often brittle and failed frequently when interfaces or services changed. The flexible nature of more reasoning capabilities being built into models allows agents to use more advanced logic and reasoning to navigate complex workflows, and to reflect and iterate until it gets it right. With agents, AI becomes both a true efficiency tool and an elevator of human capabilities for innovation, R&D, and strategic thinking.

Q: How do agents interact with AI, existing tools/services, and data?

A: An agent might depend on Large Language Models (LLMs) as its brain or to perform different components of its skills/capabilities. LLMs help the agent plan and reason, translate between structured and unstructured inputs and outputs from APIs (Application Programming Interfaces), and act as a knowledge source to supplement domain-specific data. Agents often utilize both LLMs and tools/services because it is a powerful combination of reasoning and natural language skills from the LLM, combined with precision, reliability, and access to real-world actions that are exposed through APIs.

To make this easier and more standardized with the increase in agent development, MCP (Model Context Protocol) has emerged as an increasingly adopted standard. MCP is a protocol (like an API spec) that lets tools declare their capabilities and lets the LLM call them safely and predictably. It adds the standardized interface between the LLM and those external tools/services/data. This standardization layer allows for better interoperability, scalability, and reliability.

One of the important pieces done years ago as part of our digital transformation efforts that provided an advantage for FactSet in our agentic journey—and this is applicable to other firms, too—is how we’ve taken our services, capabilities, and data and encapsulated them into micro-services exposed through APIs. With the emergence of MCP, these APIs can easily be adapted to MCP servers, providing the ease of use and standardization described above.

Q: What is the significance of agent interoperability?

A: The exciting potential of agents is to bring efficiencies to the industry by taking manual work from the user and bringing new insights at scale. The highest performance will come from highly specialized agents trained to be an expert in their task or field of data. The efficiency factor will come from these expertly trained agents working together and coordinating to replace end-to-end complex workflows. There is still work to do in defining standards for how agents interact in a positive, governed, and well-defined manner, but we believe that will unlock quite a bit of potential for firms.

In addition to agent interop, to reach a more efficient state of automation we need to understand the triggers in workflows that cause someone to act. Today, we often depend on the user to come in and kick off a process, start a workflow, or ask a question. To further elevate efficiency, firms should think about what changes in market data or signals do their employees act on and start using these changes or events to kick off agent action. FactSet’s Signals Product and API has been a useful tool both in the workstation and delivered programmatically to push market moving events, predictive machine-learning signals, and key shifts in data for different research, due diligence, and investment processes that can now start to move from humans to agents more autonomously.

Q: How should firms think about data and accuracy leading up to agents?

A: Clean, well-documented, and annotated data is the single biggest dependency for agentic AI success. Today with the GenAI tools, if the data is wrong, maybe you get an incorrect insight. With AI agents, wrong data leads to wrong actions (and on a large scale given the autonomy).

Firms need to be at a place where the data they provide to agents is in a state that allows LLMs to make the right informed decisions. Ruthlessly addressing and improving any data-quality issues, describing data and tagging data in detail with rich metadata, investing in systems that make data programmatically available to agent systems at scale, and then establishing ways to audit back to the source will be important in this whole process. Treat data like critical infrastructure.

 

Kate Stepp is Chief Technology Officer at FactSet. In this role, she leads FactSet 's technology organization and oversees its digital transformation strategy. She joined FactSet in 2008 and previously served as Senior Director of Product Management within FactSet's Research and Advisory Workflow Solutions business. Prior to that role, she was Senior Director of Engineering within FactSet 's Research Workflow Solutions business. Ms. Stepp earned a B.S. in Computer Science from Carnegie Mellon University.


This blog post is for informational purposes only. The information contained in this blog post is not legal, tax, or investment advice. FactSet does not endorse or recommend any investments and assumes no liability for any consequence relating directly or indirectly to any action or inaction taken based on the information contained in this article.

Comments

The information contained in this article is not investment advice. FactSet does not endorse or recommend any investments and assumes no liability for any consequence relating directly or indirectly to any action or inaction taken based on the information contained in this article.